The present invention is directed to a device, a method and a computer program.
Modern driver assistance systems and automated driving require a precise determination of the vehicle movement. Inertial sensor systems, wheel signals, and surroundings sensors are utilized in this case. If a video camera is installed in the vehicle, the inherent movement of the vehicle may be highly precisely estimated on the basis of the shifting of distinctive points in the image. Assuming rigid, unmoved surroundings, the movement of the vehicle is inferred in this way.
One conventional method is distinguished by the fact that it is particularly robust even in the event of breaches of the assumption of a rigid scene, for example by other moving vehicles. The reason therefor is the use of a robust estimation method, for example, the so-called random sample consensus (RANSAC), with the aid of which a hypothesis for the movement is determined, which is supported by a preferably large number of distinctive image points. One further similar method is described in U.S. Published Patent Application No. 2009/0263009 A1, in which RANSAC is utilized for outlier suppression in the movement estimation. In recent years, methods for the semantic segmentation of image information, for example, with the aid of deep learning, have also made considerable advances.